ChaoticEconomist/Jazz-Blues-Music-Dataset_SFT-or-LoRA
Viewer β’ Updated β’ 1.22k
Anuraag Rath
ChaoticEconomist
AI & ML interests
Agentic Orchestration, Fine Tuning, Reinforcement Learning, Game Theoretic Models
Recent Activity
reacted to kanaria007's post with β€οΈ about 6 hours ago
β
Article highlight: *From LLM Wrappers to SIL: A Migration Cookbook* (art-60-067, v0.1)
TL;DR:
This article is a practical migration path from todayβs LLM-heavy systems to governed SI-Core operation.
The key move is simple: stop letting the LLM *effect* the world directly. First make it a proposal engine. Then add an effect ledger, rollback, goal objects, auditable prompts, and finally move critical logic into SIL. No rewrite requiredβjust a safer commit path, step by step.
Read:
https://huggingface.co/datasets/kanaria007/agi-structural-intelligence-protocols/blob/main/article/60-supplements/art-60-067-from-llm-wrappers-to-sil.md
Why it matters:
β’ gives a realistic migration path for teams that cannot stop shipping
β’ separates LLM proposal from runtime commit authority
β’ shows how governance can be added incrementally before full SIL adoption
β’ turns βwrap the model and hopeβ into phased risk reduction
Whatβs inside:
β’ a phase map from raw LLM ops β proposal engine β effect ledger β goal objects β prompt auditability β SIL-native core
β’ typed *ActionBundle* outputs instead of direct tool calls
β’ effect-ledger + rollback-floor patterns with idempotency and compensators
β’ goal objects that make objectives computable instead of hidden in prompts
β’ guidance on where SIL pays rent first: policy gates, routing, budgets, rollback planning, and commit checks
Key idea:
Do not migrate by rewriting everything.
Migrate by moving **commit authority** first, then gradually moving **logic** into structured, replayable, checkable forms.
*LLMs can keep proposing. SI-Core must decide what is admissible to commit.* reacted to omarkamali's post with π₯ about 6 hours ago
Just sharing a little breakthrough with Gherbal LID where we managed to distinguish the 15 variants of Arabic with even higher accuracy than before, practically distinguishing Moroccan and Algerian (which overlap massively)
It also embraces the duality of MSA and arabic variants pioneered in ALDi by @AMR-KELEG et al.
Now we're only bottlenecked by the availability of high quality data for the low scoring variants such as Iraqi, Libyan, Sudanese, Adeni ...
More on Gherbal at:
https://omneitylabs.com/models/gherbal
Organizations
